Robust incentive mechanisms for efficient and secure coded edge intelligence

• Main Author: Ng, Jer Shyuan
• Other Authors: Miao Chun Yan
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/164596
• Institution: Nanyang Technological University

Coupled with reliable wireless communication technologies, the IoT devices can serve as important sources of sensor data for data-driven Artificial Intelligence (AI) applications. Instead of a single cloud server handling the large-scale datasets individually, distributed edge computing emerges as a more viable solution. The confluence of edge computing and AI technologies is known as \emph{edge intelligence}. To mitigate the challenges of distributed computing, coded distributed computing (CDC), i.e., a combination of coding theoretic techniques and distributed computing, has been proposed. In particular, CDC systems reduce communication load and alleviate the straggler effects. The implementation of CDC schemes in the distributed edge computing networks is essential in providing us with regular streams of valuable insights in the delivery of experiences, products and services. While the implementation of CDC schemes in the distributed edge computing networks is envisioned to enable a scalable and efficient edge intelligence environment that captures insights provided by the large number of heterogeneous edge nodes, the willingness to participate in the CDC tasks and the amount of resources (computation, communication, data, etc.) differ among the edge nodes. In this thesis, I first introduce the concept of CDC and provide a comprehensive survey of the development and state-of-the-art of CDC schemes. Although coding techniques have shown their effectiveness in solving the challenges of traditional distributed computing systems, there are several challenges that need to be addressed for large-scale implementations over the distributed edge computing networks that involves millions of heterogeneous edge nodes. One of the important issues is the incentive mismatch among the different entities in the networks. The different entities aim to maximize their own utilities and thus may not facilitate the completion of the CDC tasks. In consideration of the heterogeneity of the edge nodes, i.e., computational and communication capabilities as well as storage capacities, I propose incentive mechanism designs to encourage the edge nodes to facilitate the CDC tasks over the distributed edge computing networks. One of the research works is the implementation of a double auction mechanism in the coded vehicular edge computing network. While the double auction mechanism matches the edge servers with the required resources to the vehicles and determines the prices that the vehicles need to pay for the resources of the edge servers, the simulation results show that the double auction mechanism also satisfies the properties of individual rationality, incentive compatibility and budget-balance. In addition, I propose a Serverless Hierarchical Federated Learning (SHFL) framework that capitalizes the benefits of improving the reliability of the Federated Learning (FL) network and reducing the risk of single point of failure. To improve the network efficiency, I introduce a reputation aware hedonic coalition formation scheme to model the cluster formation in the SHFL network. Simulation results show that the FL workers make decisions to maximize their own utilities, regardless of the effect of their decisions on other FL workers. The hedonic coalition formation algorithm converges to a Nash-stable partition. By considering the resource inefficiency and utilization rate of the various edge nodes, effective resource allocation frameworks and incentive mechanisms can be designed to ensure successful implementation of the coding schemes in practical distributed edge computing networks. Finally, I discuss several research directions for future works.